An Intelligent Hybrid Experimental-Based Deep Learning Algorithm for Tomato-Sorting Controllers

Abstract

Conventionally, the methods used for the sorting of tomatoes are manual. These methods are costly, non-productive, and their reliability is uncertain. With advancing technology, deep-learning, and artificial intelligence techniques are being utilized to develop fully automated system controllers. The primary reason behind using these techniques is their competitive performance in solving high nonlinear classification problems. Therefore, this paper investigates the performance and combination scenarios of a number of effective artificial intelligence techniques and strategies. Improving the classification accuracy of automated tomato-sorting controllers shall also be explored. Convolution neural network (CNN), artificial neural network (ANN), self-organizing map (SOM), learning vector quantization (LVQ), and support vector machine (SVM) are developed, optimized, assessed, and compared. In this paper, three main categories are considered, namely, unripe, ripe, and defective (overripe and rotten). Moreover, an experimental setup is designed, manufactured, and tested to verify the computational results obtained from the neural networks as well as to assess the real-time performance of the proposed algorithm. As per the research findings, the utilization of a hybrid CNN-ANN-based algorithm is favored, as it demonstrated a superior performance during validation and experimental testing. The CNN-ANN-based control algorithm yielded a theoretical classification performance of 100% for all classes while the experimental results produced 100% for unripe and ripe classifications and 90% for the ripe and defective (overripe and rotten) classifications. The results of this paper have the potential to improve the classification accuracy of similar fruit and vegetable sorting machines.